Summary
Unsloth AI builds open-source tools for fine-tuning and training LLMs using custom GPU kernels and optimized math. Founded by Daniel Han, who previously worked at NVIDIA optimizing algorithms, and Michael Han, they make LLM training 30x faster with 90% less memory than Flash Attention 2. They support RL, LoRA, and full fine-tuning across models like Llama, Mistral, and Gemma. With over 10 million monthly model downloads and 40K GitHub stars, they help teams train custom models in days instead of weeks.
